LLR SUM(INPUTS) != SUM(OUTPUTS) error

jbristow · 2007-08-10, 04:59

This isn't strictly about Prime95, but I'm not sure where the best place to post this is. I'm getting a repeatble SUM(INPUTS) != SUM(OUTPUTS) error in LLR. I've attached a mix of what's showing up in the program window and what is showing up in the lresults file. I get a similar error in one of the same two bits every time I try to test these two numbers.

Can anyone reproduce this? How serious is this error? Is the program being too aggressive with FFT size here?

This is version 3.7.1b for Windows off of Jean's website (the Help-About says 3.7.0)

Thanks.

Starting Proth prime test of 271701370441215*2^380+1
Using zero-padded FFT length 48, a = 7

Bit: 32/427, ERROR: SUM(INPUTS) != SUM(OUTPUTS), 3161673192960.942 != 3161673192960.94
Possible hardware failure, consult the readme file.
Continuing from last save file.

Starting Proth prime test of 271701370441215*2^380+1
Using zero-padded FFT length 48, a = 7

Disregard last error. Result is reproducible and thus not a hardware problem.
Bit: 33/427, ERROR: SUM(INPUTS) != SUM(OUTPUTS), 741770356199.0127 != 741770356199.0127
Possible hardware failure, consult the readme file.
Continuing from last save file.

Starting Proth prime test of 271701370441215*2^382+1
Using zero-padded FFT length 48, a = 7

Bit: 32/429, ERROR: SUM(INPUTS) != SUM(OUTPUTS), 7396090213939.41 != 7396090213939.408
Possible hardware failure, consult the readme file.
Continuing from last save file.

Bit: 32/429, ERROR: SUM(INPUTS) != SUM(OUTPUTS), 14054985136451.11 != 14054985136451.11
Possible hardware failure, consult the readme file.
Continuing from last save file.

Cruelty · 2007-08-10, 08:56

I've run this test and at n=380 got similar error, after 5-minutes pause LLR resumed without error.
Attached find input and result file if anyone wishes to doublecheck this error.

Prime95 · 2007-08-13, 14:31

I did not have any problem using prime95 25.3 PRP test.

However, a PRP test is different than an LLR test. The key difference (as far as the underlying math routines are concerned) is the constant multiplier. PRP sometimes multiplies a squaring result by 3. If LLR is multiplying by a larger number, and the gwnum routines are incorrectly adjusting the error checking function for the larger constant multiplier, then you'll get the error you reported.

Is it possible to have Jean Penne look into this?

paulunderwood · 2007-08-13, 20:15

Works ok with the line "CpuSupportsSSE2=0" in llr.ini

I hope this helps Jean to find the bug.

Has anyone e-mailed him?

jbristow · 2007-08-14, 04:07

I sent Jean an email about it this weekend.

2007-08-10, 04:59	#1
jbristow Aug 2007 1011101₂ Posts	LLR SUM(INPUTS) != SUM(OUTPUTS) error This isn't strictly about Prime95, but I'm not sure where the best place to post this is. I'm getting a repeatble SUM(INPUTS) != SUM(OUTPUTS) error in LLR. I've attached a mix of what's showing up in the program window and what is showing up in the lresults file. I get a similar error in one of the same two bits every time I try to test these two numbers. Can anyone reproduce this? How serious is this error? Is the program being too aggressive with FFT size here? This is version 3.7.1b for Windows off of Jean's website (the Help-About says 3.7.0) Thanks. Starting Proth prime test of 2717013704412152^380+1 Using zero-padded FFT length 48, a = 7 Bit: 32/427, ERROR: SUM(INPUTS) != SUM(OUTPUTS), 3161673192960.942 != 3161673192960.94 Possible hardware failure, consult the readme file. Continuing from last save file. Starting Proth prime test of 2717013704412152^380+1 Using zero-padded FFT length 48, a = 7 Disregard last error. Result is reproducible and thus not a hardware problem. Bit: 33/427, ERROR: SUM(INPUTS) != SUM(OUTPUTS), 741770356199.0127 != 741770356199.0127 Possible hardware failure, consult the readme file. Continuing from last save file. Starting Proth prime test of 271701370441215*2^382+1 Using zero-padded FFT length 48, a = 7 Bit: 32/429, ERROR: SUM(INPUTS) != SUM(OUTPUTS), 7396090213939.41 != 7396090213939.408 Possible hardware failure, consult the readme file. Continuing from last save file. Bit: 32/429, ERROR: SUM(INPUTS) != SUM(OUTPUTS), 14054985136451.11 != 14054985136451.11 Possible hardware failure, consult the readme file. Continuing from last save file.

2007-08-13, 20:15	#4
paulunderwood Sep 2002 Database er0rr 3,739 Posts	Works ok with the line "CpuSupportsSSE2=0" in llr.ini I hope this helps Jean to find the bug. Has anyone e-mailed him? Last fiddled with by paulunderwood on 2007-08-13 at 20:41

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
ERROR: SUM(INPUTS) != SUM(OUTPUTS)	AES	PrimeNet	6	2007-10-05 18:48
v. 2.13: SUM(INPUTS) != SUM(OUTPUTS)	Kaiw	Software	7	2005-10-26 14:49
SUM(INPUTS) != SUM(OUTPUTS)	rudi_m	Software	3	2005-08-22 15:42
ERROR: SUM(INPUTS) != SUM(OUTPUTS)	flava	Hardware	3	2004-01-19 17:52
ERROR: SUM(INPUTS) != SUM(OUTPUTS)	ebx	Software	5	2004-01-02 22:25

2007-08-13, 14:31	#3
Prime95 P90 years forever! Aug 2002 Yeehaw, FL 2×53×71 Posts	I did not have any problem using prime95 25.3 PRP test. However, a PRP test is different than an LLR test. The key difference (as far as the underlying math routines are concerned) is the constant multiplier. PRP sometimes multiplies a squaring result by 3. If LLR is multiplying by a larger number, and the gwnum routines are incorrectly adjusting the error checking function for the larger constant multiplier, then you'll get the error you reported. Is it possible to have Jean Penne look into this?

2007-08-14, 04:07	#5
jbristow Aug 2007 3·31 Posts	I sent Jean an email about it this weekend.